Munich Personal RePEc Archive
What are the Health Benefits of a
Constant Water Supply? Evidence from London, 1860-1910
Troesken, Werner and Tynan, Nicola and Yang, Yuanxiaoyue
University of Pittsburgh, Dickinson College, University of Copenhagen
May 2021
Online at https://mpra.ub.uni-muenchen.de/107872/
MPRA Paper No. 107872, posted 21 May 2021 13:33 UTC
What are the Health Benefits of a Constant Water Supply?
Evidence from London, 1860-1910*
Werner Troeskena Nicola Tynanb
Yuanxiaoyue (Artemis) Yangc
aWerner Troesken, University of Pittsburgh, played a guiding role on this paper until his death in September 2018. To honor his work and his memory, we have retained him as a co-author. Any errors in the paper are attributable to the other authors.
b Corresponding author. Dickinson College, P.O. Box 1773, Carlisle PA 17013, USA, tynann@dickinson.edu
c University of Copenhagen, Øster Farimagsgade 5 DK-1353 Copenhagen K, Denmark, yangyuanxiaoyue@gmail.com
May 2021
*Revised version forthcoming in Explorations in Economic History (doi:
10.1016/j.eeh.2021.101402) Abstract
What are the benefits of moving from intermittent water delivery (which limits user access to less than 24 hours per day) to constant service? To address this question, we study the transition from intermittent to constant water supply in London. Between 1871 and 1910, the proportion of London households with access to a constant water supply (24 hours a day, 7 days a week) rose from less than 20 to 100 percent. Idiosyncratic delays in the negotiation process between companies and property owners generated random variation in the timing of the transition across London districts. Exploiting this variation, we find that a one percentage point increase in a local population with access to constant service decreased deaths from waterborne diseases by as much as 0.5 percent and explains approximately a fifth of the late nineteenth century decline in waterborne disease mortality. Results are robust to the inclusion of controls for population density, concerns regarding the reporting of cause-of-death, district-specific time trends, and spatial autocorrelation. In placebo tests, we find no evidence that the extension of constant service affected mortality from non-water borne diseases or deaths from violence.
Keywords: Water supply; Mortality; London, 19th century; Public health JEL classification: N33, O18, I18, N93, L95
Declarations of interest: none
1. Introduction
Around 300 million people in the world today have only intermittent access to piped water supplies. In many cities in Asia, Africa, and Latin America, urban residents receive piped water for only a few hours every day or have access to piped water only two or three days per week (Lee and Schwab 2005: 114; Kumpel and Nelson 2016; Heymans et al. 2016: 11). While there is broad consensus that disruptions in water access pose a serious public health risk, due to pipe intrusion or water contamination during storage, and that universal constant service is desirable, systematic assessments of the health benefits of constant service, particularly in large metropolitan areas, are a nascent focus of development research (Lee and Schwab 2005; Ercumen et al. 2015; Galaitsi et al. 2016; World Bank Group 2017). Yet the existing literature in economics still focuses almost exclusively on discrete interventions related to water filtration and chlorination. Accordingly, in this paper, we ask: what are the mortality effects of moving from an intermittent water supply (IWS) to a constant water supply (CWS)? To address this question, we study the transition from IWS to CWS in London during the late nineteenth and early twentieth century. Between 1871 and 1910, the proportion of London households with access to CWS (24 hours a day, 7 days a week) rose from less than 20 to nearly 100 percent.
London is a promising setting for three reasons. First, the introduction of CWS necessitated non-trivial investments on the part of both water companies and private home owners, which could only be implemented after extensive multi-party negotiations. Idiosyncratic delays in the negotiation process generated plausibly exogenous variation in the timing of the transition from IWS to CWS across London districts. Second, while the transition from IWS to CWS was a widespread phenomenon in Western cities during the late nineteenth and early twentieth century, London’s transition is well documented, particularly in relation to district level data on access and mortality. Third, one might worry that cities were adopting other technological innovations to
improve water quality at the same time they introduced CWS. In London, however, the most rapid growth in the extension of constant service came decades after a series of legal and regulatory interventions designed to protect the London water supply from contamination, and by the same token, the city’s constant, high pressure, distribution network was largely complete when the city municipalized the system in 1902 (Hardy 1991: 85-92; Metropolitan Water Board 1953).
The analysis below proceeds as follows. First, we convert data on mortality for all of London’s registration districts for the period from 1860 to 1910 into mortality data for 20 composite health districts to address district boundary changes during the period. We then estimate the proportion of homes in each of these districts with access to CWS using data on connection rates across London’s private water companies and the relative presence of these companies across districts. Although most companies and districts do not exhibit sharp increases in the provision of CWS until after 1875 and by 1910 there is near universal access to CWS, the data suggest meaningful differences in access to CWS across both districts and time in the intervening years.
Exploiting this variation, we adopt a difference-in-differences strategy to identify how access to CWS affected waterborne disease mortality.
The results suggest that a one percentage point increase in a local population with access to CWS decreased deaths from waterborne diseases by as much as 0.5 percent. We show that these results are robust to controls for population density and district-specific time trends and concerns regarding low quality reporting of cause-of-death during the early years of our sample. Recent work in economic history raises the possibility that improvements in health-related infrastructure might induce in-migration to treated places. If so, calculated mortality rates based on interpolated population data (as ours) might be biased downward for inter-census years. Although this is less likely to be a concern for constant service than with more discrete sanitary interventions, we follow
the extant literature and implement remedial tests that show our results are unaffected by such concerns. Finally, in placebo tests, we find no evidence that the extension of CWS affected disease or other causes of death that were not correlated with access to CWS,using all deaths minus water- borne disease deaths, whooping cough (an airborne disease), and deaths from violence.
Aside from the implications for access to CWS in the developing world today, there is now a vast literature in economics exploring how access to water affects a wide range of economic and health outcomes (e.g., Cutler and Miller 2005;Günther and Fink 2010; Beach, Ferrie, Saavedra and Troesken 2016; Knutsson 2017; Alsan and Goldin 2019; Anderson, Charles and Rees 2019;
Gallardo-Albarrán 2020). This literature, however, focuses almost exclusively on discrete interventions related to changes in water source, chlorination, or filtration. To our knowledge, we are the first paper in the economics literature to establish and measure the causal link between constant service and mortality. This linkage is significant on at least two levels. First, it suggests an important complementarity that has yet to gain recognition in the economics literature on the health effects of water. More precisely, the results here show that improved water supplies, protected from organic and inorganic pathogens, is necessary but not sufficient for promoting health; centralized purification measures are relatively ineffective if service and access are inconsistent and vulnerable to regular interruptions or water is contaminated after treatment.
Second, according to the estimated coefficients here, introducing CWS had large mortality effects, on par in magnitude with other, more heavily studied, interventions and explaining as much as 40% of the decline in waterborne disease mortality in London between 1871 and 1910.
2. London’s Water System: A Brief History
In this section, we briefly review the relevant history of the London water system. The first part of this narrative focuses on water treatment; the central message here is that by 1856, most
London households had access to relatively safe piped water from one of the eight metropolitan water companies. The second part of the narrative focuses on the transition from IWS to CWS and shows that this transition was concentrated in the twenty-five year interval between 1875 and 1900, and occurred during a period where other water related interventions were infrequent and less significant than those that occurred before 1860.
Table 1 provides a compact summary of the water treatment and protection strategies adopted by London’s eight water companies. As can be seen from the table, four of the eight companies withdrew their water from the River Thames and another three withdrew their water from the River Lee, a major tributary of the Thames. Only one company (Kent) relied on deep wells for its supply. The Thames supplied approximately half of London’s water, with the River Lea and groundwater sources providing another quarter each. The two most common strategies to guarantee the purity and safety of river water were to either install filtration systems or move water intakes upstream to prevent sewage from contaminating the supply. Some of London’s water companies began investing in water treatment and protection during the early 1800s, before the advent of regulatory mandates. For example, the Chelsea Water Works Company started to filter its water in 1829 and the Lambeth Water Works built a filtration plant in 1841. In 1852, the Lambeth Water Works also moved its water intake upriver and opened a new waterworks at Seething Wells.The Kent Company introduced sand filtration in 1845, but abandoned filtration when it began supplying only groundwater from local wells in 1861.
Table 1: London Water Companies, Water Sources and Treatment Company Water source First filtration Move
upriver
Adopt Waterworks Clauses Act
Chelsea River Lea 1829 1856 1852 & 1853
East London River Lea 1854 1834 1852
Grand Junction River Thames 1855 1855 1852
Kent Deep wells 1845 (until 1861) n/a 1864
Lambeth River Thames 1841 1852 1848
New River River Lea;
wells 1855 Abandoned
Thames 1852 1852
Southwark & Vauxhall River Thames 1855 1855 1852
West Middlesex River Thames 1855 1855 1852
This table shows water source, opening date of first filtration plant, date abstraction location moved away from central London, and date the Waterworks Clauses Act 1847 – the first to mention constant service – was adopted for each of London’s eight water companies.
Sources: Information in this table was drawn from Francis Bolton, London Water Supply (1884) and Arthur Shadwell, The London Water Supply (1899).
Although some companies were investing in water treatment on their own, the Metropolis Water Act of 1852 provided further impetus. The Act required all London water companies to follow industry best practice by moving surface water intakes upstream and constructing filtration plants. The Act also stipulated that by 1856 all reservoirs within a five-mile radius of St. Paul’s Cathedral had to be covered unless stored water was subsequently filtered and, by December 1855, all surface water had to be filtered and supplied to customers only in covered pipes or aqueducts.
All water companies met the targets established by the Metropolis Water Act 1 and, by 1856, customers in all but a few areas were receiving filtered water. The next major step in water purification measures (as opposed to distribution measures such as constant service) did not occur
1 For example, in the wake of the Metropolis Act, the Lambeth Company added larger filter beds when it moved its intake upriver, as did the Chelsea Company in 1856. The other three companies supplying water from the river Thames built new filter beds when they moved their intakes upriver in 1855. The East London Company, supplying water from the river Lea, added filter beds in 1854, while the New River Company delivering water from the river Lea, and from springs and wells via the New River, added filtration in 1855. The East London Company had abandoned its intake at Old Ford, within tidal reach, in 1834. The New River Company already withdrew water from the river Lea above the tidal reach and stopped withdrawals from the river Thames in 1852. (See Bolton 1884)
until 1916 with the introduction of chlorination (Jones 2012: 105-121; Metropolitan Water Board 1953).
In 1866, cholera broke out in parts of a district served by the East London Company. This outbreak resulted from an illegal decision to connect the company’s supply lines to an uncovered reservoir and spurred two investigations into metropolitan water supply. The first investigation, in 1866, focused on assessing the impact of the Metropolis Water Act, 1852. The second, made by a Royal Commission appointed in 1867, assessed higher grounds in England and Wales as potential sources of water in addition to the overall state of water supply in the metropolis. The subsequent report found existing supplies satisfactory in terms of quantity and quality. However, it found that the one area where the companies failed to meet the expectations of the 1852 Act was in regards to frequency of supply (i.e., it was concerned that IWS remained the norm in London) and here it recommended new legislation to compel the introduction and extension of constant service across all London districts.
London officials were promoting CWS as early as the 1840s. This was reflected in both the writings of Edwin Chadwick (1842:48) and non-binding legislation passed in 1847, which, in turn, helped prompt the East London Water Company to begin providing CWS to some customers in its district. The Company required the removal of domestic water storage cisterns for those receiving CWS to prevent in-house contamination of water.2 The goal of constant high-pressure service was restated in the Metropolis Water Act, 1852, but with sufficient caveats that it proved ineffective. Parliament continued to investigate and debate the need for a constant supply of water at high-pressure, culminating in passage of the Metropolis Water Act of 1871. Giving clear
2Removing cisterns was thought to provide an additional guarantee against on-site water contamination after the move to CWS. The downside of removing cisterns was that it left customers without water during any interruption of service. After 1876, most companies recommended the retention of cisterns to protect against unforeseen disruption of supply during periods of drought or frost.
legislative mandates, the 1871 Act fostered a commitment to introduce and extend CWS among all eight water companies. As a result, the proportion of houses in London receiving water 24 hours a day, 7 days a week, increased from less than 20% in 1871 to almost 100% in 1910.
Genuinely constant service – 24-hours, seven days a week – required direct connection to a main or service pipe constantly charged with water at high pressure. Maintaining a constantly high pressure in water pipes prevented the intrusion of contaminated groundwater or water from leaking sewers. However, it increased the risk of waste – in today’s language “unaccounted for water” – either in transit or due to inappropriate in-house plumbing. Water providers had legal control over leaks in the network, but almost no control of domestic waste; they could not require homeowners to fix leaks or assure that their plumbing could withstand the high pressure that came with CWS. In this setting, IWS was a second-best solution that gave water providers control over water availability and system pressure, particularly during emergencies such as fires or drought (Hardy 1991; Hillier 2014). This was one reason the 1871 Metropolis Water Act was so important:
it gave companies greater control to enact and enforce the regulations on household plumbing necessary for CWS to work at a system wide level.
Hence, before committing themselves to CWS and making the requisite investments (which could be sizeable), water companies wanted homeowners to coordinate and make their own investments in improving their household plumbing systems. As early as 1851, Thomas Wicksteed (an engineer for the East London Company) argued that the main barriers to CWS were the companies’ lack of power to check houses had appropriate fittings to prevent waste and landlords’
unwillingness to pay for necessary plumbing. Reluctance on the part of homeowners to make these investments persisted well into the 1880s. For example, the 11th annual report of the Local Government Board noted that resistance to the extension of the constant service system came from
owners and occupiers reluctant to incur the additional expense (Local Government Board, 1882:
cxxiv). In the wake of the 1871 Act, water companies could overcome such reluctance through litigation, though that could be a costly and time-consuming path.3
The Grand Junction Company’s process for introducing CWS serves to illustrate the underlying coordination problems. For all new houses, the company’s inspector determined that pipes and fittings met the regulations and passed approval to the surveyor who connected the house to the company’s main. For older houses, changing from intermittent to constant service, owners or occupiers had to remove waste pipes from cisterns, substituting warning pipes and adding cisterns to the lead communication pipes before connecting to the main (Bolton 1881:15). The difficulties facing the Grand Junction Company (as well as other London water companies) increased with the arrival of new housing developments at higher elevations outside the city.
Providing CWS to these areas required water delivered at higher pressure. An 1881 Parliamentary investigation led by Frank Bolton, Water Examiner appointed under the 1871 Metropolis Water Act, concluded that the only way to assure such pressure was to give companies greater authority to regulate and enforce rules governing household plumbing and preventing waste of water.
3 In November 1883, the Southwark & Vauxhall Company summoned J. McDonald, landlord and owner of several houses on Orb Street, Newington, to appear before a magistrate for failing to install appropriate plumbing and wasting between 5,000 and 6,000 gallons per day from one house. The judge fined him a surprisingly large sum of
£10 on the grounds that “it was highly necessary the work of introducing the constant supply should be carried out.”
"At LAMBETH, Mr. J. M'DONALD, owner of several houses." Times [London, England] 6 Nov. 1883: 3. The Times Digital Archive. Web. 24 Apr. 2018.
Figure 1: Proportion of London homes with CWS, 1860-1910
The early estimate for CWS is calculated using linear interpolation between 1847 and 1875; only the East London Water Company rolled out CWS during this period.
Sources: Annual Reports of the Local Government Board, 1876-1910.
After the Metropolis Water Act of 1871, the two most important legislative measures affecting the provision of public service were the creation of the London County Council (LCC) in 1889 and the Metropolitan Water Board (MWB) in 1902. The LCC had the power to compel water companies to introduce CWS in the older areas of London, where there was greater reluctance to adopt constant service. This measure, however, appears to have done little more than nudge companies to continue a trend that began after the 1871 measure. This can be seen in Figure
1, which plots the proportion of London homes on CWS from 1860 to 1910.4 In 1904, the MWB took over management of London’s eight water companies. The MWB completed the transition from IWS to CWS, particularly for houses in South London districts supplied by the Lambeth Company - the one company with less than 95% of houses on constant service in 1904.5 By 1910, fewer than one percent of houses did not receive their water directly from the mains with constant service.
3. Constant Service and Waterborne Diseases
The central question we ask here is if the extension of CWS between 1860 and 1910 reduced waterborne disease rates. Motivating our analysis, Figure 2 plots the natural log of the death rate from all waterborne diseases (i.e., typhoid, cholera, diarrhea, and dysentery) in London against the proportion of all homes in the metropolis without CWS, so still receiving water intermittently. The graph reveals three patterns. First, CWS expands rapidly in the wake of the 1871 Metropolis Water Act and slows after 1898 as near universal service is reached. Second, waterborne disease rates are rising in the years before the Metropolis Water Act and the onset of more rapid growth in CWS. Third, aside from the sudden dip in mortality after 1905, waterborne disease rates stop falling after 1895 and the slowdown in the expansion of constant service once near universal access is reached. Between 1871 and 1894, waterborne disease mortality was halved as the proportion of London households with CWS increased from less than 20% to 80%.6
4Data on the number and percentage of houses supplied with CWS by each water company are taken from the Annual Reports of the Local Government Board starting in 1876. Prior to 1876, data on CWS is only available for a few years; we use linear interpolation to estimate missing years.
5The percentage of houses on constant service for the New River Company experiences a small drop in 1905 after MWB acquisition but returns to more than 98% within two years. This is likely a result of the expansion of ‘water London’ to a larger area under the MWB and to our method of calculating CWS coverage using company totals.
6 The increase in water-related deaths between 1895 and 1900 may be attributable to drought years and temporary reversion to IWS by the East London Company from 15 July-18 October 1895, 17 July-19 September 1896, and 23 August-7 December 1898 (London Metropolitan Archives, ACC/2558/MW/C/15/25). The heatwave itself may have altered the disease environment and contributed to this increase in mortality (Hanlon, Hansen and Kantor 2020: 5).
Figure 2: Proportion of homes without CWS and waterborne disease mortality
Sources: Registrar General’s Annual Reports for 1860-1910 (23rd-73rd; 1862-1912) corrected by Graham Mooney, The Johns Hopkins University; Annual Reports of the Local Government Board, 1876-1910.
The literature identifies two primary sources of post-treatment contamination of water delivered intermittently.7 Low pressure in distribution pipes allows intrusion of contaminants, particularly in crowded areas where water pipes lie in close proximity to sewers (Kumpel and Nelson 2014: 2770; Lee and Schwab 2005: 115).8 A comparison of CWS and IWS in eight pairs
There is evidence of food-borne typhoid from oysters and other shellfish during the late 1800s and shellfish regulation explains some of the steep drop in mortality after 1905 (Hardy 2014).
7See Lee and Schwab (2005) for other sources of contamination. As they point out, it is often a combination of system flaws, exacerbated under IWS, that contribute to reduced water quality.
8Switching from IWS to CWS often allowed companies to identify system leaks and replace or repair pipes.
of matched wards in India found that 99.3% of CWS water samples met standards requiring the absence of E. coli while only 68.3% of IWS samples met the standards (Kumpel and Nelson 2013:
5). Water stored in cisterns or other containers by consumers to ensure a constant domestic supply is another source of contamination, partly just from sitting, often uncovered, between supply periods and partly as a result of consumer neglect. Water stored by consumers for more than a day has tested for significantly higher levels of E. coli than water delivered continuously (Kumpel and Nelson 2013: 11, 13). In his 1884 report to Parliament, Frank Bolton identified cisterns as a source of water contamination under IWS in London.9
In the analysis that follows, we use district-level data to explore the possibility that the correlations between IWS and mortality observed in Figure 2 reflect a causal relationship. London had 37 distinct registration districts in 1860 but only 31 in 1910 and the district borders changed somewhat over time. To address this boundary change issue, we have created composite health districts with time consistent borders.10 Those borders, for all twenty districts, are depicted in the map in Figure 3. Each of these health districts was served by one or more of London’s eight water companies and varied both in terms of the evolution of their waterborne disease rates over time and the pace at which they introduced constant service. Using a difference-in-differences strategy, we exploit this variation in disease and access to CWS across both districts and time to estimate the causal impact of extending CWS on mortality.11
9 “The Water Companies are frequently blamed for delivering unpotable water, when if the true delinquent were sought it would be found to be the water consumer himself, whose lack of attention to his cisterns and filters has created the evil of which he complains.” (Bolton 1884: 11)
10Mortality data kindly provided by Graham Mooney included codes for the creation of these composite health districts.
11 Appendix Figure A1 reproduces Figure 2 for each health district.
Figure 3: Composite London Health Districts
Sources: Graham Mooney, The Johns Hopkins University; Great Britain Historical GIS Project.
Our full analysis focuses on the years from 1860 to 1910. Ideally, we would have direct counts of the number of households with CWS (i.e., receiving water 24 hours a day, 7 days a week) across districts and over time. Unfortunately, those data do not exist. Instead, we estimate the proportion of homes in every district with CWS by building up from company level data on the following: (1) the number households served by each company with CWS; and (2) market penetration for each company at the district level. As briefly alluded to above, before 1904, eight private companies supplied water through their own networks, and this division of water
infrastructure continued after 1904, during the first few years of management by the Metropolitan Water Board. These eight companies had statutory approval to supply certain parishes, often with more than one company supplying parts of the same district. Of our 20 composite London districts, five received water from one company, ten from two companies, four from three companies, and only Greenwich received water through four distinct networks.12
We know the population supplied by each water company within each registration district in 1904. These data are from Appendix A of the First Annual Report of the Metropolitan Water Board (Metropolitan Water Board 1905: 17-29). We use these data to estimate the population share connected to each water company’s network in those districts supplied by more than one company.
The estimates are reported in Table 2. We treat these proportions as fixed over time, in part because the statutory boundaries of the water companies make it unlikely that the proportion of the population served by each company changed substantially during the period. Construction and demolition of houses will have introduced small changes, but these are unlikely to have been concentrated so heavily in one company’s part of a district to have introduced big changes in population shares.13
12 In some cases, the statutory boundaries overlapped and either two companies supplied the same parish or they reached boundary agreements (Local Government Board, 1891: 325). For a full list of parishes each company was authorized to supply, see Metropolitan Water Board, 1908: 92-111.
13Following slum removal schemes in the last quarter of the nineteenth century, most tenants of demolished homes moved only short distances (Yelling 1986).
Table 2: Percent of London’s 20 Composite Health Districts Supplied by Each Water Company
Health District \
Company Chelsea East London
Grand
Junction Kent Lambeth New River
Southwark
&
Vauxhall
West Middlesex
Bethnal Green 100
Camberwell 13 20 67
Chelsea 100
Greenwich 1 86 11 2
Hackney 50 40
Hampstead 19 81
Holborn 100
Islington 100
Kensington 21 33 46
Lambeth 63 37
London City 1.5 98.5
Marylebone 9 91
Pancras 85 15
Poplar 100
Shoreditch 25 75
Southwark 52 48
St. Olave Southwark
4 1 95
Stepney 97 3
Wandsworth 34 66
Westminster 58 24 18
Each row represents a composite health district and the numbers in each cell represent the percent of that district supplied by the water company named in the column heading. All rows sum to 100.
Source: First Annual Report of the Metropolitan Water Board, 1905.
We then gather data on constant service for each company from the Annual Reports of the Local Government Board (LGB). The LGB annual reports give the number and percentage of houses on CWS and the total number of houses supplied by each company for every year between 1876 and 1906. After 1906, the reports provide updates for those companies not yet providing CWS throughout their network. Only the East London Company provided CWS to a significant number of households before 1875. The company had introduced constant service early, in 1847, and had over 50% of customers on the mains in 1875. By 1904, all except the Lambeth Company had over 95% of customers receiving CWS. By 1910, every Metropolitan Water Board customer, except a few at high elevation in the district previously supplied by the Lambeth Company,
received water 24-hours a day. Figure 4 plots the proportion of households served by each company with CWS over time.14
Figure 4: Percent of Houses Supplied by Each Company on CWS, 1860-1910 Sources: Annual Reports of the Local Government Board, 1876-1910.
We combine the data on water company penetration at the district level (Table 2) and the company level data on CWS (Figure 4) to estimate district-level measures of the proportion of homes with access to CWS. To do this, we multiply the share of a composite health district
14Data for East London Company before 1875 is an interpolation between the values for 1847 and 1875.
supplied by each company by the percentage of the population supplied by that company on CWS.
Generally, our measures of CWS for any district i in year t can be written as, 𝐷𝑖𝑠𝑡𝑟𝑖𝑐𝑡𝐶𝑊𝑆𝑖𝑡 = ∑ 𝜃𝑖𝑘∙ (𝐶𝑜𝑚𝑝𝑎𝑛𝑦𝐶𝑊𝑆𝑘𝑡)
𝑛
𝑘=1
where, n is the number of companies providing water to district i, θik is the proportion of the population in district i that is consuming water supplied by company k, and CompanyCWSkt is the proportion of water company k’s customers who enjoy CWS in year t.15 By this measure, all time- series variation in district level CWS is derived from company level changes in the number of customers with CWS.
Figure 5 shows the variation in CWS across districts over time. While all districts exhibit increases in access to CWS after the Metropolis Water Act of 1871, there is meaningful cross- district variation in both rate of increase and the magnitude of the increase. Some districts (e.g., Chelsea and St. Olave) concentrate the roll out of CWS over a short window of only a few years, while other districts spread out the introduction of CWS over many years (e.g, Greenwich, Islington, and Southwark) or even decades (Hackney, Shoreditch, and Stepney). In addition, because some districts (e.g., Bethnal Green and Poplar) begin the study period with relatively high levels of initial access to CWS, districts also vary in terms of the magnitude of the increase in access to CWS.By 1910, all districts, except a few with customers on the old Lambeth Company network, have made the transition to CWS.
15 Consider, for example, our measure of district CWS (DistrictCWS) for the district of Camberwell. Camberwell was served by three companies: Kent, Lambeth, and Southwark & Vauxhall. Given the levels of market penetration for each of these companies, we calculate the level of CWS in the Camberwell district as follows: DistrictCWS- Camberwell = 0.13 (Kent)t + 0.2 (Lambeth)t + 0.67 (Southwark&Vauxhall)t. Kent is the percentage of Kent water company customers on CWS in year t, and 0.13 is the percentage of the population of Camberwell supplied by the Kent Company; Lambeth is the percentage of Lambeth water company customers on CWS and 0.2 is the percentage of the population of Camberwell supplied by the Lambeth Company; and so on for the Southwark & Vauxhall company.
Figure 5: Percent of Houses in Each Composite Health District with CWS, 1870-1910 Sources: Registrar General’s Annual Reports for 1860-1910 (23rd-73rd; 1862-1912) corrected by Graham Mooney, The Johns Hopkins University; Annual Reports of the Local Government Board, 1876-1910, Appendix A of the First Annual Report of the Metropolitan Water Board
In our formal empirical analysis, we identify the impact of CWS after controlling for district and year fixed effects, district-specific time trends, and other potential confounding factors. There are two keys to the analysis. First, the extent to which districts differ in the pace, timing, and magnitude at which they introduce CWS, as demonstrated in Figure 5. Second is the extent to which we can treat these differences as randomly assigned.
As explained above, the rate at which companies moved towards CWS was determined partly by the willingness of customers to install appropriate plumbing and partly by a company’s installing pipes able to withstand the pressure. The inter-district variation observed in Figure 5 was
therefore driven by idiosyncratic delays in the negotiation process between companies and property owners. Requests to convert to CWS were made at the parish, vestry or neighborhood level so not every part of a district, or even a parish, made the transition simultaneously. Following the 1871 Metropolis Water Act, companies initiating a conversion to CWS were required to publish their intention at least three months in advance to give all customers affected time to install appropriate plumbing. Advertisements in The London Gazette show that company-initiated transitions to CWS within the same district took place over multiple years. For example, the New River Company announced its intention to transition parts of St. Pancras to CWS in 1883, 1884, 1885, 1886 and 1887, though even by 1887 only parts of St. Pancras had made the transition (Hardy 1991: 87).16 For customer initiated requests, Hillier (2014: 232-3) argues that homeowners varied in their enthusiasm for modernization and some parish councils resisted constant service due to cost, disruption, or its perception as an imposition of central power on local governments.17, 18
Willingness to convert to CWS might be expected to correlate with customer wealth, either because wealthier customers were more likely to own their own home so avoid any principal-agent problem or because companies supplying wealthier districts would have expected customers to pay the cost of installing appropriate plumbing. This raises a concern that customer wealth will confound with CWS’ effect on mortality because wealthier people would have had better access to health services so be less likely to die from digestive diseases. As a result of stronger support for CWS by the East London Company, however, some poor East London districts were the first
16 These announcements appeared in The London Gazette on 16 November 1883 (issue 25288, page 5444); 11 July 1884 (issue 25375, page 3183); 7 October 1884 (issue 25402, page 4375); 3 April 1885 (issue 25457, page 1541);
16 April 1886 (issue 25578, page 1845); and 11 March 1887 (issue 25682, page 1462).
17 Fittings for constant service cost approximately £8 in the 1860s (Hillier 2011: 47), although they had fallen to £5 by the late 1880s (Hardy 1991: 78). Using the measuringworth.com purchasing power calculator, that translates to approximately £500 today.
18 In Appendix B, we calculate a counter-factual measure of CWS that assumes each company systematically extends constant service starting with its largest district. The results are much weaker with a smaller coefficient and lower significance.
to convert and this resulted in a negative metropolis-wide relationship between the expansion of CWS and wealth. To confirm this relationship between CWS and district wealth, we calculate a poverty index measure for 19 of our 20 composite districts (all except Woolwich) using a version of Charles Booth’s geographic, household-level, measure of social class in London in 1896, digitized and georeferenced by Scott Orford (Orford et al. 2002).19 Using OLS, we regress our measure of district level CWS on this poverty index, the interaction between the poverty index and a time trend, plus year fixed effects. Table 3 reports the results.
Table 3: CWS and Poverty
Constant service
Poverty index 0.677***
(0.0639) Time trend * Poverty index -0.714***
(0.0667)
Year FE Y
Observations 700
R-squared 0.819
Robust standard errors in parentheses
*** p<0.01, ** p<0.05, * p<0.1
Sources: Constant service data comes from the Annual Reports of the Local Government Board and Appendix A of the First Annual Report of the Metropolitan Water Board. Poverty data was provided by Scott Orford and is described in Orford et al. (2002).
19We used the same methodology as employed in Orford at al. (2002). This approach converts Booth’s seven classes into five classes that correspond with the Registrar General’s five social classes; a higher index number represents greater poverty.The coverage of Booth’s poverty map is less extensive than the geography of London so some districts have poverty data only for the edge of the district closest to central London.
Early expansion of CWS in the East London districts of Stepney and Poplar, as shown in Figure 5, meant poorer neighborhoods were first to receive a constant supply of water.20 Table 3 reinforces this, showing that districts with a less wealthy population on average were more likely to have converted from IWS to CWS.21 The negative coefficient on the time-trend and poverty index interaction term shows that the relationship between poverty and CWS diminished over time as all districts moved towards universal CWS. Therefore, we believe that district wealth does not challenge the impact of CWS on mortality even though we cannot use our time-invariant measure of poverty in our empirical analysis because it will be absorbed by our district fixed effects.
4. Empirical Strategy and Results
We use a generalized difference-in-differences approach in which exogenous improvements in water quality, as a result of a neighborhood switch from IWS to CWS, reduces waterborne disease mortality within a district. Our baseline model is
𝑚𝑜𝑟𝑡𝑖𝑡 = 𝛼 + 𝛽𝐷𝑖𝑠𝑡𝑟𝑖𝑐𝑡𝐶𝑊𝑆𝑖𝑡+ 𝑿𝒊𝒕′𝜹 + 𝜆𝑡+ 𝜂𝑖+ 𝛾𝑖𝑡+ 𝜀𝑖𝑡
where mortit is the natural log of deaths per 10,000 from all waterborne diseases (cholera, typhoid, diarrhea, and dysentery) for district i in year t. DistrictCWSit is the estimated proportion of homes in district i in year t that receive CWS; 𝑿𝒊𝒕 represents a vector of control variables; λt, ηi, and γit represent year fixed-effects, district fixed-effects, and linear district-year trends; and ε is a random error term. We estimate this model using ordinary least squares (OLS) with district-level mortality
20 William Booth Bryan, engineer to the East London Company confirmed this in his testimony before the 1892 Royal Commission on Metropolitan Water Supply stating that the approximately 2,000 houses not on constant service were in the wealthiest part of the company’s district in Buckhurst Hill and Woodford.
21In his 1884 report, Frank Bolton was quite critical of some wealthier residents with IWS who placed cisterns on the top floors of their house to increase water pressure, putting them out of sight and making them more difficult to clean (Bolton 1884).
data for the period 1860-1910. Graham Mooney provided the mortality data from the Registrar General’s Annual Reports for 1860-1910 (23rd-73rd; 1862-1912) but corrected for hospital deaths.22 For nineteenth-century London, official mortality statistics are often distorted by a failure to return hospital and workhouse deaths to person’s home district (Mooney, Luckin and Tanner 1999; Hardy 1993). Such misallocation of deaths in institutions could be high depending on the institution, district, and cause of death (Mooney, Luckin and Tanner 1999: 239-241). Mooney’s corrected data allows us to improve upon other studies using raw data for London during the nineteenth century. The control variables in 𝑿𝒊𝒕 include measures of population density and non- water borne disease mortality, calculated simply using all deaths minus water-borne disease deaths. Summary statistics for all variables are available in appendix Table A1.
Regression results for our core model are presented in Table 5. All regressions include year and district fixed effects. We report robust standard errors and wild bootstrap p-values. We have too few districts to use clustered standard errors without correction so, to account for within cluster (district) correlation, we correct the inference with a wild bootstrap method (Cameron et al. 2008) and report wild bootstrap p-values obtained from the boottest command in Stata described by Roodman (2015).23
22 The Registrar General reports have data on total and infant mortality by sub-district and by age and cause of death for some years. However, consistent time series data by cause of death is not available at a lower geographic level than district.
23We also computed standard errors that correct for spatial correlation, serial correlation and heteroskedasticity using code developed by Solomon Hsiang (Hsiang, 2010). The spatial correlation used weights of 1.6km and 3.5km to capture the proximity of small, central London districts but a greater distance between the centroids of larger districts on the periphery. The results are very similar to those reported here.
Table 5: CWS and Water-Borne Disease Mortality
Log of Waterborne Mortality
(1) (2) (3) (4) (5)
Constant service -0.610*** -0.443*** -0.601*** -0.573*** -0.337***
(0.0689) (0.0850) (0.0674) (0.0682) (0.0720)
[0.00] [0.00] [0.00] [0.00] [0.00]
Population density 0.00274*** 0.00450*** 0.00281***
(0.000654) (0.00160) (0.000677) [0.0003] [0.0118] [0.00]
Non-water mortality 0.00462*** 0.00390*** 0.00394***
(0.000833) (0.00112) (0.000776)
[0.00] [0.0002] [0.00]
Year FE Y Y Y Y Y
District FE Y Y Y Y Y
District time trend N Y N N N
Time period 1860-1910 1860-1910 1860-1910 1876-1910 1860-1903
Observations 1,020 1,020 1,020 700 880
R-squared 0.813 0.875 0.828 0.851 0.788
Robust standard errors in parentheses; Wild bootstrap p-values in brackets
*** p<0.01, ** p<0.05, * p<0.1
Column 1 shows the impact of the move away from IWS on our broader measure of waterborne disease mortality. The coefficient on CWS is statistically significant at the 1% level and shows that a one percentage point increase in the population of a district receiving water 24- hours a day, 7 days a week, reduced waterborne disease mortality in the district by 0.6%.24
Column 2 adds a district specific time trend. This lowers the impact of a one percentage point increase in CWS on waterborne disease mortality to a reduction of only 0.4% but it remains
24Using typhoid alone gives similar results but with a smaller coefficient and lower significance. This makes sense given our transmission mechanism through pipe infiltration and water contamination during storage, compared to studies focusing on large-scale transmission of the typhoid bacteria throughout a city. Results are available upon request.
highly significant. Column 3 adds measures of population density and non-waterborne disease mortality to the baseline regression. Both variables have the expected signs: waterborne disease mortality is higher in more densely populated districts and districts with higher water-related deaths are less healthy places overall. CWS remains significant, suggesting that we are not picking up some other factor reducing multiple types of mortality within a district at the same time as companies roll out CWS.
As robustness checks, we add columns 4 and 5, which repeat the regression in column 3 for two different time periods. To remove the time period with lower quality reporting of cause of death and of customers with access to CWS, we add column 4, starting in 1876, the year when all companies start to report the share of their customers on constant service and after more accurate reporting of waterborne disease deaths with the clear distinction of typhoid and typhus after 1869.
To avoid any unobserved effect of the introduction of the Metropolitan Water Board on mortality, we add column 5, which ends in 1903, just before the transfer of water supply to the Metropolitan Water Board. When we restrict our time period to 1860-1903, the impact of a one percentage point increase in CWS on water-borne disease mortality falls to 0.34%. Otherwise, the results are similar to our preferred regression in column 3, implying that an expansion of CWS from less than 20%
to 100% of the population explains between 27% and 40% of the reduction in water-borne disease mortality between 1860 and 1910.
5. Sensitivity Analysis
5.1 Narrowing the Window
Our baseline analysis covered the whole period 1860-1910 to fully capture the impact of the long transition from intermittent to constant water supply. Table 5, columns 4 and 5, narrowed
the window to begin with the publication of CWS data in the annual reports of the Local Government Board and to end with the transfer of London’s water infrastructure to the Metropolitan Water Board. In Table 6 we report results for different time periods to focus on the period during which most expansion of CWS took place and to account for other possible concerns relating to our choice of years.
Columns 1-4 report results starting in 1871, the year the Metropolis Water Act was passed into law and two years after separate reporting of typhoid and typhus deaths. Because the transition to CWS was nearly complete by the end of the nineteenth century, and slum clearance takes off in the 20th century, we end the period in 1900. The results in columns 1-4 confirm that CWS played an important role in reducing water borne disease mortality in London. While the coefficients are smaller than those in table 5, with the impact of a one percentage point increase in CWS on waterborne disease mortality falling to closer to 0.2%, CWS retains its significance.
Table 6: CWS and Water-Borne Disease Mortality Log of Waterborne Mortality
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Constant service -0.254*** -0.181** -0.237*** -0.252*** -0.203** -0.221** -0.255** -0.203*** -0.233**
(0.0802) (0.0890) (0.0770) (0.0911) (0.0933) (0.0907) (0.117) (0.0770) (0.0993) [0.0021] [0.0432] [0.0024] [0.0061] [0.0368] [0.0181] [0.0307] [0.0097] [0.0198]
Population density 0.005*** -0.009** 0.0056*** 0.0079** 0.0197*** 0.008*** 0.0122***
(0.001) (0.004) (0.0013) (0.0032) (0.0051) (0.0018) (0.0025)
[0] [0] [0.0492] [0.0881] [0] [0] [0.0279]
Non-water mortality 0.0041*** 0.0014* 0.0043*** 0.0042*** 0.0064*** 0.0051*** 0.0059***
(0.0008) (0.0008) (0.00086) (0.0013) (0.0015) (0.001) (0.0011) [0.0028] [0.0001] [0.0001] [0.0001] [0] [0.0001] [0]
Year FE Y Y Y Y Y Y Y Y Y
District FE Y Y Y Y Y Y Y Y Y
District time trend N Y N Y N N N N N
Time period 1871-1900 1871-1900 1871-1900 1871-1900 1871-1900 1882-1900 1882-1894 1876-1900 1876-1894
Missing years None None None None 1895, 1896,
1898
None None None None
Observations 600 600 600 600 540 380 260 500 380
R-squared 0.813 0.872 0.832 0.876 0.836 0.835 0.863 0.823 0.849
Robust standard errors in parentheses; Wild bootstrap p-values in brackets
*** p<0.01, ** p<0.05, * p<0.1
As noted above, the East London Company temporarily reverted to IWS in some neighborhoods during the summer of 1895, 1896 and 1898. This means our measure of CWS for districts supplied by the East London Company may be overstated for these years; results in column 5 exclude them. The droughts that diminished the East London Company’s water supply were a result of heatwaves in London during the late 1890s. Recent research has shown that these heatwaves may have changed London’s disease environment in ways that increased digestive disease mortality (Hanlon, Hansen and Kantor 2020).25 Comparing columns 4 and 5 suggests expansion of CWS in other companies’ districts slowed the increase in mortality during the drought years. Columns 7 and 9 show results for a period ending in 1894, before the drought years and temporary reversion to IWS. Columns 6-9 also narrow the window to focus on the period after 1876, when the Local Government Board started reporting annual CWS data, and after 1882, during which most expansion of CWS took place. Constant service remains significant in all specifications providing strong support for our thesis that CWS played in important role in reducing deaths from waterborne disease.
5.2 Addressing concerns about population extrapolation
We do not have particular concerns about migration between districts during this time period. District level wealth is highly persistent, changing little between the late 1800s and 1991, and there is no evidence to suggest shifts in the geography of poverty during our time period (Orford et al. 2002). Slum clearance was limited until the early 20th century, resulting mostly from rail or road construction and accelerating only slowly after the Housing of the Working Classes
25 Hanlon, Hansen and Kantor (2020) argue that hot summers after 1894 likely slowed the reduction in infant mortality by five years. Although diarrhea is an important cause of infant death and a number of studies have shown that improvements in water quality and sanitation have reduced infant mortality (e.g., Knutsen 2015, Alsan and Goldin 2019), the climate impacts present a challenge for our analysis. Nevertheless, we show results using infant mortality as our dependent variable in Appendix A.
Act of 1890. Most people who moved as a result of housing destruction before 1900 remained within the same district, often increasing crowding in remaining buildings (Yelling 1986;
Hobhouse 1994).
Table 7: CWS and Water-Borne Disease Deaths
Waterborne Deaths
(1) (2) (3) (4) (5)
Constant service -37.47** -105.9** -74.21*** -69.98*** -75.35***
(17.75) (45.77) (14.10) (13.32) (21.98)
[0.0259] [0.0122] [0.00] [0.00] [0.0001]
Population density -0.641* -0.327 -0.552
(0.327) (0.278) (0.359) [0.0288] [0.258] [0.1133]
Non-water deaths 0.0614*** 0.0553*** 0.0623***
(0.00772) (0.00508) (0.00838)
[0.00] [0.00] [0.00]
Year FE Y Y Y Y Y
District FE Y Y Y Y Y
District time trend N Y N N N
Time period 1860-1910 1860-1910 1860-1910 1876-1910 1860-1903
Observations 1,020 1,020 1,020 700 880
R-squared 0.601 0.682 0.678 0.891 0.654
Robust standard errors in parentheses; Wild bootstrap p-values in brackets
*** p<0.01, ** p<0.05, * p<0.1
Nevertheless, our mortality measures are rates that rely on census population data and interpolation between census years. There might be concerns that this hides annual population changes. To address this, we repeat our regressions from Table 5 using waterborne disease deaths as our dependent variable and non-waterborne deaths as a control. The results are presented in Table 7. Using disease deaths rather than mortality rates does not change the results. CWS remains
highly significant in all specifications. London districts with more people receiving reliable, high- pressure, water delivery 24-hours a day, 7 days a week, reported lower deaths from water-borne diseases than those with lower rates of CWS.
5.3 Thresholds in the transition to CWS
Intermittent water supply allowed for local contamination of previously treated water through pipe infiltration or water sitting in uncovered cisterns. Transitioning a neighborhood to constant service removed these sources of local contamination, reducing the spread of waterborne disease. In testing this connection, we use a continuous measure of CWS coverage; our CWS coefficient shows the impact of a one percentage point increase in CWS whether that takes a district from zero to 1% or from 99% to 100%. If transitions to CWS generated positive neighborhood spillovers or there were negative spillovers from neighborhoods remaining on IWS, there may be threshold levels of CWS coverage that were required to reduce mortality. To test this, we divide our continuous measure into bins for each 10% increase in CWS coverage, with a baseline of 0-10% of the population in a district receiving water constantly under pressure.
Figure 6 plots the resulting coefficients and standard errors using our baseline model with year and district fixed effects and controls for population density and non-waterborne disease mortality. It shows the marginal effect of moving another 10% of a district from intermittent to constant service. The results suggest that our baseline continuous model is reasonable. Other than the move from 50-60% to 60-70% coverage, every threshold increase in connections reduces mortality.26
26 The regression results for this and other specifications parallel to those in Table 5 are available upon request.
Figure 6: Marginal effects of each 10% increase in CWS coverage
The figure shows the marginal effects of percentile increases in CWS coverage with 95% confidence intervals.
Sources: Registrar General’s Annual Reports for 1860-1910 (23rd-73rd; 1862-1912) corrected by Graham Mooney, The Johns Hopkins University; Annual Reports of the Local Government Board, 1876-1910.
5.4 Placebo Tests
To check that we are not picking up some other source of mortality decline that correlates with the district-level roll-out of CWS, we re-run our baseline regressions using three alternative measures of mortality as our dependent variable. The first is the measure of non-waterborne disease mortality, defined above as overall minus waterborne disease mortality. The second is deaths from violence, a cause that has no obvious relationship with water quality. The third is whooping cough, an air-borne disease that had a particularly high mortality rate in London and declined rapidly during the second half of the nineteenth century (Hardy 1993: 10-11). Our measure of non-waterborne disease mortality is the same variable used in our vector of controls.
Data for deaths from violence and whooping cough comes from the Registrar General’s Annual Reports for 1860-1910 (23rd-73rd; 1862-1912).27
We use violence because finding a good disease as a placebo test of the impact of CWS on mortality is not easy. One challenge is the role that the increased availability and reliability of water plays in improving the broader sanitary environment and reduced mortality from non- waterborne diseases (Ferrie and Troesken 2008). Efforts to reduce the transmission of air-borne diseases are aided by improvements in personal hygiene and sanitation, including hand washing.
Most literature on water-washed or water-related diseases focuses on roundworm, hookworm, conjunctivitis, scabies and other infections found in developing country or refugee settings but recent evidence suggests that handwashing also reduces the risk of acute respiratory infections (Cairncross 2003).28 For late nineteenth century London, the availability of water likely resulted
27 Transcribed registration district mortality data was kindly shared with us by Brian Beach. We mapped registration district data to our 20 health districts.
28 Cairncross (2003) argues that handwashing may provide explanation for the Mills-Reincke phenomenon observed during the late nineteenth century (Ferrie and Troesken 2008).
in more frequent handwashing and may have played a role in reducing the transmission of a number of non-waterborne diseases through these broader improvements in the sanitary environment.
The second challenge for finding good placebo tests is the increasing frequency of admission to London hospitals in one district of infectious disease patients who lived in another district. These deaths were reported in the district housing the hospital rather than the patient’s own residential district. This concern is particularly problematic for smallpox, scarlet fever, and diphtheria. As Anne Hardy (2013: 300) notes, “the arrival of the MAB hospitals from the 1870s, and especially with the dramatic rise in hospitalization of cases of scarlet fever, diphtheria, and typhoid after 1891, district mortality figures for these diseases become increasingly doubtful; by the 1890s they are often meaningless. The clearest example of this fallacy in the registration data comes with smallpox.” Whooping cough does not suffer from this challenge because London hospitals did not accept whooping cough patients (Hardy 1993: 23).29
29 Even though raw typhoid data suffers from the hospitalization fallacy, our measure of waterborne disease mortality does not because we use corrected data provided by Graham Mooney.
